OWASP LLM Top 10: A Legacy Archive That Points Elsewhere

The Repository Has Been Retired as an Entry Point

The README carries an IMPORTANT block that states the repository is maintained as a legacy entry point and historical archive, and that active development has moved to GenAI-Security-Project/GenAI-LLM-Top10. That single sentence changes how the repository should be read. It is not a codebase you clone and run. It is the former home of the OWASP project page, kept online so that existing links, citations, and historical artifacts continue to work. The README says exactly that. New issues, pull requests, and release work are directed to the active repository. The last push date on the default branch is 2026-08-05, which is consistent with a repository that still receives pointer updates even though its content is frozen. If you arrived here from a search engine or an old blog post, the material you are looking at may be one or two editions behind the current guidance.

What the Archive Actually Contains

Three releases are listed. The 2024 tag is labelled OWASP Top 10 for Large Language Model Applications 2025 and is dated 2024-11-18. Two older tags, 2023-v1.1 and 2023-v1, are dated 2024-11-11. The README also links to an Archive directory described as holding earlier releases and translations, and to the 2025 release as an archived item. So the repository functions as a document store: tagged PDFs and source files, translated editions, and working files. The primary language is listed as Python, which is worth noting because it sets an expectation the README does not support. Nothing in the supplied material describes an installable package, a CLI, or a library. The Python label most likely reflects tooling used to build or process the publication, not a runtime dependency you would add to a project.

Where the Current Guidance Lives

The README points to the OWASP GenAI LLM Top 10 2026, published August 4, 2026, hosted at genai.owasp.org, with source browsing under the 2026/final path in the GenAI-LLM-Top10 repository. Corrections go through a release-errata issue template, and broader feedback through a release-feedback template. The project is described as a core initiative within the OWASP GenAI Security Project, which the README says develops freely available guidance and resources for securing generative AI systems. This matters for anyone building a threat model. If your model references an entry by its identifier, that identifier may have been renumbered, merged, or dropped between the 2023, 2025, and 2026 editions. The archive cannot tell you that. The errata and release pages in the active repository can.

There Is No Run Step for This Repository

The honest answer to how you get this running is that you do not. The README gives no install command, no requirements file, no configuration keys, and no entry point. What it gives is a set of URLs. The current release lives at the genai.owasp.org resource page for the 2026 edition. The source tree is browsable at the 2026/final path in GenAI-Security-Project/GenAI-LLM-Top10. Corrections are filed through the release-errata issue template in that repository, and pull requests go to the same place. The only operational surface described here is the pages-build-deployment GitHub Actions workflow, whose badge sits at the top of the README and which builds the project page from the main branch. If you are looking for something to pip install, this is the wrong repository and the wrong project.

The Practical Failure Mode Is a Stale Citation

The way this archive causes harm is quiet. An engineer finds a well-written 2023 entry, quotes its identifier in a design document or a compliance checklist, and ships it. The 2026 edition exists, but the archive does not flag which entries changed. Nothing in the supplied material lists a diff between editions, and the README does not promise one. A second failure mode is contribution. Someone files an issue or opens a pull request here, and the README states plainly that new issues, pull requests, and release work should be directed to the active repository. Effort spent here is likely to be redirected rather than merged. A third case is tooling. Because the primary language field says Python, an automated dependency scanner may treat this repository as a software component. It is not one, and the scanner output will be noise.

Compared with NIST AI RMF and MITRE ATLAS

The closest alternatives are not competitors to this repository so much as different instruments. The NIST AI Risk Management Framework is organised around governance functions and an organisational risk process rather than a numbered list of application-layer weaknesses, so it answers how to run an AI risk programme, not which prompt-injection entry to cite. MITRE ATLAS is built as an adversary technique knowledge base mapped to observed tactics, which makes it useful for red-team planning and detection engineering, while the OWASP list is organised as an awareness and prioritisation document for application teams. The practical difference is granularity and audience. If you need a shared vocabulary for a design review, the OWASP list is the shorter read. If you need to model an attacker's sequence of steps, ATLAS is structured for that and the OWASP list is not.

Licence, Reuse, and What ShareAlike Means for Derived Work

The README states the project is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License, and the badge repeats CC BY-SA 4.0. The repository metadata field reports NOASSERTION, which is a scanner result rather than a licence statement, and the README is the clearer source here. Two obligations follow from BY-SA. Attribution is required, and derivative works must carry the same licence. For a team that wants to paste entries into an internal wiki, that is usually fine. For a team that wants to embed the text in a proprietary product or a training course sold under restrictive terms, the share-alike condition is the thing to review with counsel. I am not giving legal advice, and the exact boundary of what counts as a derivative work is a question for a lawyer, not for a README.

Maintenance Cost Is Now Close to Zero, and So Is the Return

There is nothing to upgrade here. The repository receives pointer and badge updates, and the README explicitly tells readers where the work moved. The cost of keeping it in a citation list is that a reader following your link lands on an archive notice and has to make one more hop. The cost of not checking that hop is a stale identifier in a document that other people rely on. The maintenance burden sits entirely on the active repository, where release errata and feedback are filed. The OWASP Slack channel #team-genai-top-10-llm is named in the README for working group discussion, which is the route to take if you want to raise something about the current edition rather than the archived one.

Editorial conclusion

Use this repository only to resolve old citations, retrieve archived translations, or inspect the 2023 and 2025 editions. Anyone starting new work should go to GenAI-Security-Project/GenAI-LLM-Top10 and the 2026 publication linked from the README. Before citing anything from this archive, open the 2026 release page and check whether the entry still exists under the same identifier.

Official sources

1. Issues
2. OWASP/www-project-top-10-for-large-language-model-applications on GitHub
3. Project website
4. README
5. Releases